Publication number  US6788646 B1 
Publication type  Grant 
Application number  US 09/559,988 
Publication date  Sep 7, 2004 
Filing date  Apr 26, 2000 
Priority date  Oct 14, 1999 
Fee status  Paid 
Publication number  09559988, 559988, US 6788646 B1, US 6788646B1, USB16788646, US6788646 B1, US6788646B1 
Inventors  Gabor Fodor, Miklós Telek, Sándor Rácz 
Original Assignee  Telefonaktiebolaget Lm Ericsson (Publ) 
Export Citation  BiBTeX, EndNote, RefMan 
Patent Citations (16), NonPatent Citations (37), Referenced by (55), Classifications (6), Legal Events (5)  
External Links: USPTO, USPTO Assignment, Espacenet  
This Application for Patent claims the benefit of priority from, and hereby incorporates by reference the entire disclosure of, U.S. Provisional Application for Patent Serial No. 60/159,351, filed Oct. 14, 1999.
The present invention generally relates to the field of communication networks, and in particular to link capacity sharing and link bandwidth allocation in such networks.
Many communication networks of today support socalled elastic traffic such as the “best effort” services provided in Internet Protocol (IP) based networks or the Available Bit Rate (ABR) traffic in ATM networks. Elastic traffic is typically established for the transfer of a digital object, such as a data file, a Web page or a video clip for local playback, which can be transmitted at any rate up to the limit imposed by the link capacity. Web browsing on the Internet in particular is a good and representative example of elastic traffic. Here, the “elasticity” of the traffic is apparent as the userperceived throughput (normally given in transmitted bits or bytes per time unit) when downloading for example a web page fluctuates in time depending on the overall system load.
The services delivered by IP based networks and the Internet in particular are called “best effort”, because the networks generally do not provide any guarantee of the quality of service (QoS) received by the applications. The IP network only makes a best effort to provide the requested service. For instance, if an application requests the network to deliver an IP packet from one endpoint to another, the network normally can not say what the delay through the network will be for that packet. In fact, the network does not even guarantee tat the packet will be delivered at all.
Therefore, terminals connected to an IP network have to handle packet losses and excessive packet delay situations. Such situations occur when there are too many applications simultaneously using the network resources. These congestion situations have a nonzero probability in IP based networks, because IP networks do not exercise call admission control (CAC). In other words, IP networks do not restrict the number of simultaneously connected users, and consequently if there are too many users utilizing the network resources there will be congestion and packet losses.
However, with the advent of realtime traffic and QoS requirements in EP networks, there is a need for exercising call admission control (CAC) in order to restrict the number of connections simultaneously present in the network.
An important aspect of call or connection admission control is that new calls arriving to the network may be rejected service in order to protect inprogress calls. In general, CAC algorithms such as those commonly in use for rigid traffic in conventional ATM networks provide a basic means to control the number of users in the network, thereby ensuring that admitted users get the bandwidth necessary to provide the QoS contracted for. Consequently, a CAC algorithm represents a tradeoff between the blocking probability for new calls and the provided throughput for inprogress calls. In other words, the more users that the CAC algorithm admits into the network (which reduces the blocking probability) the smaller the provided throughput peruser becomes, since a greater number of users will share the total bandwidth, and vice versa.
Recent research has indicated that it is meaningful to exercise call admission control even for elastic traffic, because CAC algorithms provide a means to prevent TCP sessions from excessive throughput degradations.
The issue of applying CAC for elastic connections, and thereby providing a minimum throughput for Transmission Control Protocol (TCP) connections in the Internet has been addressed by Massoulie and Roberts in references [13]. Here, bandwidth is allocated to different users according to some fairness criteria.
It has been recognized by Gibbens and Kelly in references [45] that there is an intimate relationship between throughput and blocking probabilities for elastic traffic, and that this tradeoff is connected to the issue of charging.
It has also been shown by Feng et al. in reference [6] that providing a minimum rate guarantee for elastic services is useful, because in that case the performance of the TCP protocol can be optimized.
As the Internet evolves from a packet network supporting a single best effort service class towards an integrated infrastructure for several service classes, there is also a growing interest in devising bandwidth sharing strategies, which meet the diverse needs of peakrate guaranteed services and elastic services.
Similarly, modern ATM networks need to support different service classes such as Constant Bit Rate (CBR) and Available Bit Rate (ABR) classes, and it is still an open question how to optimally share the link capacity among the different service classes.
In general, the issue of bandwidth sharing, in the context of dynamically arriving and departing traffic flows and especially when users have different throughput and blocking requirements, is known from the classical multirate circuit switched framework to be an extremely complex problem.
The present invention overcomes these and other drawbacks of the prior art arrangements.
It is a first object of to invention to devise a link capacity/bandwidth sharing strategy that meets the diverse needs of rigid and elastic services in a mixed rigidelastic traffic environment.
In particular, it is desirable to treat the issues of bandwidth sharing and blocking probabilities for elastic traffic in a common framework. In this respect, it is a second object of the present invention to provide a link capacity sharing mechanism that considers the throughputtoblocking tradeoff for elastic traffic. Specifically, it would be beneficial to develop and utilize a link capacity sharing algorithm that optimizes the throughputtoblocking tradeoff.
It is a further object of the invention to provide an appropriate calllevel model of a transmission link carrying elastic traffic and to apply the calllevel model for dimensioning the link bandwidth sharing for throughputblocking optimality.
These and other objects are met by the invention as defined by the accompanying patent claims.
The invention concerns an efficient strategy for sharing link bandwidth in a mixed rigidelastic traffic environment, as well as a strategy for sharing bandwidth among elastic traffic flows.
Briefly, the idea according to the invention is to share link capacity in a network by dividing the link capacity into a first common part for elastic as well as rigid (nonelastic) traffic and a second part dedicated for elastic traffic based on received network traffic inputs. Subsequently, one or more admission control parameters for the elastic traffic are determined based on the division of link capacity as well as received network traffic inputs.
The division of link capacity generally serves to share the link capacity between rigid and elastic traffic, and in particular to reserve a part of the link capacity to elastic traffic. Preferably, a minimum required capacity of the common part relating to rigid traffic is determined given a maximum allowed blocking probability for the rigid traffic. In this way, a certain grade of service (GoS) on call level is guaranteed for the rigid traffic on the link.
The admission control parameter(s) determined for elastic traffic generally serves to restrict the number of elastic traffic flows simultaneously present on the link. In particular, by formulating a calllevel model for elastic traffic and determining a maximum number of admissible elastic traffic flows based on calllevel constraints for the elastic traffic related to throughput and/or blocking, the throughputtoblocking tradeoff is fully considered. In this respect, the invention is capable of optimally allocating link bandwidth among elastic connections in the sense that blocking probabilities are minimized under throughput constraints, or the other way around, in the sense that the throughput is maximized under blocking constraints. In his way, the invention provides maximum link bandwidth utilization, either in terms of minimum blocking under throughput constraints or maximum throughput under blocking constraints.
Accordingly, an efficient strategy for sharing bandwidth in a mixed rigidelastic traffic environment is provided. In particular, the bandwidth sharing algorithm guarantees a maximum blocking for rigid traffic as well as a minimum throughput and/or a maximum blocking for elastic traffic.
An important technical advantage of the invention is its ability to meet the diverse needs of rigid traffic and elastic traffic.
Another advantage of the invention is the ability to provide predictable quality of service for both the user and the network provider while at the same time ensuring high network provider revenue.
By considering only the elastic traffic of the overall traffic in a mixed rigidelastic traffic environment, or alternatively by reducing the common bandwidth part to zero so that the entire link is reserved for elastic traffic, the overall link capacity sharing mechanism is reduced to We determination of one or more admission control parameters for elastic traffic. Admission control for requested new elastic connections can then be exercised based on such admission control parameter(s). In particular, by minimizing the blocking probabilities with respect to the number of admissible elastic connections under given throughput constraints for the elastic traffic, excessive blocking probabilities are avoided, while ensuring a given user throughput.
Another aspect of the invention concerns the application of a calllevel model of a link supporting elastic traffic, for dimensioning the link bandwidth sharing for throughputblocking optimality in an admissioncontrol enabled IP network. In particular, an elastic traffic flow is modeled as having a bandwidth that fluctuates between a minimum bandwidth and peak bandwidth during the holding time of the traffic flow. Furthermore, the elastic traffic is associated with at least one of a minimum accepted throughput and a maximum accepted blocking probability.
A further aspect of the invention concerns a computational method for determining a Maxkov chain steady state distribution that is particularly advantageous for large state spaces. The Markov chain describes the dynamics of a link carrying a number of traffic classes including nonadaptive elastic traffic, and the computational method provides a good initial approximation of the steady state distribution based on Markov chain product form calculations.
Other aspects or advantages of the present invention will be appreciated upon reading of the below description of the embodiments of the invention.
The invention together with further objects and advantages thereof, will be best understood by reference to the following description taken together with the accompanying drawings, in which:
FIG. 1 is a schematic diagram of a communication network according to a preferred embodiment of the invention;
FIG. 2 is a schematic flow diagram of the overall link capacity sharing algorithm applied in a mixed rigidelastic IP traffic environment according to a preferred embodiment of the invention;
FIG. 3 is a schematic block diagram of pertinent parts of an IP router according to a preferred embodiment of the invention;
FIG. 4 is a Markov chain state space diagram for an illustrative transmission link system;
FIG. 5 is a graph illustrating the mean and the variance of the throughput of adaptive elastic flows as a function of their service time for an illustrative example of a transmission link system;
FIG. 6 is a schematic diagram illustrating the elastic cutoff parameters that fulfill given QoS requirements for an illustrative example of a link system; and
FIG. 7 is a schematic flow diagram of the overall link capacity sharing algorithm for a mixed CBRABR traffic environment according to a preferred embodiment of the invention.
Throughout the drawings, the same reference characters will be used for corresponding or similar elements.
Throughout the disclosure, the terms connection and flow are used more or less interchangeably for what is traditionally denoted as a call.
For a better understanding, a general overview of an illustrative communication network according to a preferred embodiment of the invention will now be made with reference to FIG. 1.
FIG. 1 is a schematic diagram of a communication network according to a preferred embodiment of the invention. The communication network is here illustrated as an IP based network, but may be in the form of an ATM network or any other type of network or combination of networks capable of supporting elastic traffic. The communication network 10 is based on a number of interconnected IP routers 20 (ATM switches in the case of an ATM network) forming the core network. The core network is accessed by different users 30 (computers, servers, etc.) through access points, with a so called usernetwork interface (UNI) being defined for the interaction between the IP routers and the user equipment. Typically, a plurality of users are connected to some form of aggregation point, such as an access router (AR) 40, which acts an intermediate between the endusers and the core network.
Link capacity sharing, also referred to as bandwidth control in the network context, and packet scheduling normally reside on the network side of the UNI, within the IP routers 20. In particular, the bandwidth control and packet scheduling are preferably implemented at the output port side of the routers 20. The overall bandwidth control has two main functions. First, the bandwidth control serves to share the bandwidth between different traffic classes. Second, the bandwidth control serves to restrict the number of simultaneously active connections within the traffic classes. The latter function is hereinafter referred to as call admission control (CAC), and is normally exercised at the input port side of the IP routers 20, where connections are accepted or rejected in accordance with some CAC algorithm. The overall bandwidth control algorithm, including the CAC algorithm, may be for example implemented as hardware, software, firmware or any suitable combination thereof.
The usernetwork contract (UNC) is typically defined at the UNI. The UNC usually indicates the QoS to which the user is entitled and also the specification of the traffic, which the user may inject into the network, along with supplementary data. The supplementary data may include, for example, the time of day during which the user has access to a particular service, etc. For instance, the UNC may specify that no more than 1% of the userinjected IP packets (or ATM cells) may be lost by the network and that the user may send in 10 Mbits during any one second through the UNI.
The CAC part of the bandwidth control algorithm may use the UNC information of multiple users to provide a basic means to control the number of simultaneously present users in the network, thereby ensuring that the admitted users receive the bandwidth required to provide the contracted QoS. The CAC algorithm represents a tradeoff between blocking probabilities and the provided throughput; the more users that the CAC algorithm admits into the network (which reduces the blocking probability), the smaller the provided throughput per user becomes because a greater number of users will share the network bandwidth.
CAC may be realized by means of the classical signaling exchange known for example form conventional communication circuitswitched networks. However, if the majority of the elastic flows in the network are short, such as for many TCP flows on the Internet today, the introduction of classical sign exchange to perform admission control may result in large overheads. Therefore, an onthefly decision to accept or discard the first packet of a flow as suggested in reference [3] would be more beneficial. For this reason, a mechanism based for example on the existing Resource Reservation Protocol (RSVP) is provided for keeping track of the identities of currently active flows, and for classifying packets according to these identities as and when they arrive. To determine whether a flow is new or not, it is sufficient to compare its identifier with that of the flows on a special list of active flows. If no packet was received for a certain flow within a given timeout period, the flow would be removed from the list of active flows. Admission control is preferably realized by determining a maximum number of admissible flows based on the prevailing traffic situation in the system, and setting the size of the list of active flows accordingly. If the list is full, a new flow will be rejected. Otherwise, the flow will be accepted and entered into the list.
In the following, a link capacity sharing algorithm according to a preferred embodiment will be described with reference to the particular application of an IP based network carrying rigid as well as elastic traffic.
First, a proper calllevel traffic model needs to be formulated. Unfortunately, the application of the classical multirate calllevel models for management of elastic traffic, such as best effort traffic in IP networks or ABR traffic in ATM networks, is everything but straightforward. For example, it is not possible to associate elastic traffic with a constant bandwidth. Instead, the bandwidth occupied by elastic traffic flows fluctuates in time depending on the current load on the link and the scheduling and rate control algorithms applied in the network nodes. The notion of blocking, when applied to elastic traffic flows, is not as straightforward as for rigid traffic, because an arriving elastic flow might get into service even if at the arrival instant there is no bandwidth available. Besides, for many services, the actual residency time of an elastic flow depends on the throughput actually received by the elastic flow.
Multiclass Model of a Transmission Link for Mixed Rigidelastic Traffic
In the following, an example of a feasible Markovian model of a transmission link serving both peakbandwidth assured (rigid or nonelastic) and elastic traffic classes is formulated. For simplicity and clarity, only a single rigid traffic class and two elastic traffic classes are considered. It should be noted that the model as well as the accompanying link sharing algorithm can be extended to more general cases, and of course even simpler cases.
The system under consideration comprises a transmission link of capacity C, which by way of example can be regarded as an integer number in some suitable bandwidth unit, say Mbps. In this example, calls arriving at the link generally belong to one of the following three traffic classes:
Class 1—Rigid traffic class flows, characterized by their peak bandwidth requirement b_{1}, flow arrival rate λ_{1 }and departure rate μ_{1}.
Class 2—Adaptive elastic class flows, characterized by their peak bandwidth requirement b_{2}, minimum bandwidth requirement b_{2} ^{mm}, flow arrival rate λ_{2 }and departure rate μ_{2}. Although the bandwidth occupied by adaptive elastic flows may fluctuate as a function of the link load, their actual holding time is not influenced by the received throughput throughout their residency in the system. This is the case for instance with an adaptive video codec, which, in case of throughput degradation decreases the quality of tie video images and thereby occupies less bandwidth.
Class 3—Nonadaptive elastic class flows, characterized by their peak bandwidth requirement b_{3}, minimum bandwidth requirement b_{3} ^{min}, flow arrival rate λ_{3 }and ideal departure rate μ_{3}. The ideal departure rate is experienced when the peak bandwidth is available. The real instantaneous departure rate is proportional to the bandwidth of the flows.
We denote the actual bandwidth allocated (reserved) to a flow of class2 and class3 in a given system state with b_{2} ^{r }and b_{3} ^{r}, both of which vary in time as flows arrive and depart. We will also se the quantity r_{min}=b_{1} ^{min}/b_{i }(for i=2 or i=3) associated with elastic flows with minimum bandwidth requirements.
One may think of a nonadaptive elastic class flow as one that upon arrival has an associated amount of data to transmit (W) sampled from an exponentially distributed service requirement, with distribution
which in the case when the peak bandwidth b_{3 }is available during the entire duration of the flow gives rise to an exponentially distributed service time with mean 1/μ_{3}. Since the free capacity of the link fluctuates in time according to the instantaneous number of flows in service, the bandwidth given to the nonadaptive elastic flows may drop below the peak bandwidth requirement, in which case the actual holding time of the flow increases.
All three types of flows arrive according to independent Poisson processes, and the holding time for the rigid and adaptive flows are exponentially distributed. As we will see, the moments of the holding time of the nonadaptive flows can be determined using the theory of Markov reward processes. In short, two types of elastic traffic are considered. Elastic traffic is associated with both a peak and a minimum bandwidth requirement, and allowed into service only as long as the minimum bandwidth requirement is fulfilled. The two elastic traffic classes primarily differ in terms of how their residency time depends on the acquired throughput.
To ensure a given QoS of the different elastic classes (that, in general, differ in their peak and minimum bandwidth, i.e. b_{2}≠b_{3}, b_{2} ^{min}≠b_{3} ^{min}) we need to establish some policy, which generally governs the bandwidth sharing among the different elastic classes. For this reason, we define the following general bandwidth sharing rules between the elastic classes. The following presentation concerns only two elastic classes, but it extends naturally to more than two elastic classes.
If there is enough bandwidth for all flows to get their respective peak bandwidth demands, then class2 and class3 flows occupy b_{2 }and b_{3 }bandwidth units, respectively.
If there is a need for bandwidth compression, i.e. n_{1}·b_{1}+n_{2}·b_{2}+n_{3}·b_{3}>C, then the bandwidth compression of the elastic flows is such that r_{2}=r_{3}, where r_{2}=b_{2} ^{r}/b_{2 }and, r_{3}=b_{3} ^{r}/b_{3}, as long as the minimum rate constraint is met for both elastic classes (i.e. b_{2} ^{min}/b_{2}≦r_{2}≦1 and b_{3} ^{min}/b_{3}≦r_{3}≦1).
If there is still need for farther bandwidth compression, but either one of the two elastic classes does not tolerate further bandwidth decrease (i.e. r_{i }is already b_{i} ^{min}/b_{i }for either i=2 or i=3) at the time of the arrival of a new flow, then the service class which tolerates further compression decreases equally the bandwidth occupied by its flows, as long as the minimum bandwidth constraint is kept for this traffic class.
Three underlying assumptions of the above exemplary model are noteworthy. First of all, it is assumed that both types of elastic flows are greedy, in the sense that they always occupy the maximum possible bandwidth on the link, which is the smaller of their peak bandwidth requirement (b_{2 }and b_{3}, respectively) and the equal share (in the above sense) of the bandwidth left for elastic flows by the rigid flows (which will depend on the link allocation policy used). Second, it is assumed that all elastic flows in progress share proportionally equally (i.e. the r_{i}'s are equal) the available bandwidth among themselves, i.e. the newly arrived elastic flow and the inprogress elastic flows will be squeezed to the same r_{j }value. This assumption, as we will see, provides a quite “fair” resource sharing among the elastic flows. To have different elastic traffic classes with significantly different QoS this assumption needs to be modified. If a newly arriving flow decreased the elastic flow bandwidth below b_{2} ^{min }and b_{3} ^{min }(i.e. both elastic classes are compressed to their respective minima), that flow is not admitted into the system, but it is blocked and lost. Arriving rigid as well as elastic flows are allowed to “compress” the inservice elastic flows, as long as the minimum bandwidth constraints are kept. As a third point, the model assumes that the rate control of the elastic flows in progress is ideal, in the sense that an infinitesimal amount of time after any system state change (i.e. flow arrival and departure) the elastic traffic sources readjust their current bandwidth on the link. While this is clearly an idealizing assumption, the buffers at the IP packet layer could be made large enough to absorb the IP packets until TCP throttles the senders. The fact that the model assumes immediate source rate increase whenever possible make the forthcoming throughput and blocking calculations conservative rather than optimistic.
It is intuitively clear that the residency time of the nonadaptive elastic flows in this system depends not only on the amount of data they want to transmit, but also on the bandwidth they receive during their holding times, and vice versa, the amount of data transmitted through an adaptive elastic flow depends on the received bandwidth. In order to specify this relationship we define the following quantities:
θ_{2}(t) and θ_{3}(t) defines the instantaneous throughput of adaptive and nonadaptive flows, respectively, at time t. For example, if there are n_{1}, n_{2 }and n_{3 }rigid, adaptive, and nonadaptive flows, respectively, in the system at time t, the instantaneous throughputs for adaptive and nonadaptive flows are min(b_{2}, (C−n_{1}b_{1}−n_{3}r_{3}b_{3})/n_{2}) and min(b_{3}, (C−n_{1}b_{1}−n_{2}r_{2}b_{2})/n_{3}), respectively. Note that θ_{2}(t) and θ_{3}(t) are discrete random variables for any t≧0.
defines the throughput of an adaptive flow having a holding time equal to t.
{tilde over (θ)}=∫_{0} ^{∞}{tilde over (θ)}_{τ}dF(τ)=μ_{2}∫_{0} ^{∞}{tilde over (θ)}_{τ}e^{−μ} ^{ 2 } ^{τ}dτ (random variable) defines the throughput of an adaptive flow, where F(t) is the exponentially distributed holding time.
T_{x}=inf{t∫_{0} ^{τ}θ_{3}(τ)dτ≧x} (random variable) gives the time it takes for the system to transmit x amount of data through an elastic nonadaptive flow.
{circumflex over (θ)}=x/T_{x }defines the throughput of an nonadaptive flow during the transmission of x data units. Note that θ_{x }is a continuous random variable.
{circumflex over (θ)}∫_{0} ^{∞}{circumflex over (θ)}_{x}dG(x)=μ_{3}/b_{3}∫_{0} ^{∞}{circumflex over (θ)}_{x}e^{−xμ} ^{ 3 } ^{/b} ^{ 3 }dx (random variable) defines the throughput of an nonadaptive flow, where the amount of transmitted data is exponentially distributed with parameter μ_{3}/b_{3}.
Although a number of general bandwidth sharing rules have been defined above, a more specific link capacity sharing policy, especially one that considers the diverse requirements of rigid and elastic traffic, still needs to be presented.
Link Capacity Sharing Algorithm
According to the invention, the Partial Overlap (POL) link allocation policy, known from reference [7] describing the POL policy for simulative analysis in the classical multirate circuit switched framework, is adopted and modified for a traffic environment that includes elastic traffic. According to the new so called elastic POL policy, the link capacity C is divided into two parts, a common part C_{COM }for rigid as well as elastic traffic and a dedicated part C_{ELA}, which is reserved for the elastic flows only, such that C=C_{COM}+C_{ELA}.
Furthermore, admission control parameters, one for each elastic traffic class present in the system, arc introduced into the new elastic POL policy. In this particular example, N_{EL2 }denotes the admission control parameter for adaptive elastic flows and N_{EL3 }denotes the admission control parameter for nonadaptive elastic flows. Each admission control parameter stands for the maximum number of admissible flows of the corresponding elastic traffic class. The admission control parameters are also referred to as cutoff parameters, since as long as the maximum number of simultaneous elastic flows of a certain class are present on the link, new elastic flows will be rejected, a form of cutoff.
Under the considered elastic POL policy, the number (n_{1}, n_{2}, n_{3}) of flows in progress on the link is subject to the following constraints:
In (1) the elastic flows are protected from rigid flows. In (24) the maximum number of elastic flows is limited by three constraints. Expression (2) protects rigid flows from elastic flows, while (34) protect the inprogress elastic flows from arriving elastic flows. The new elastic POL policy is fully determined by the division of the link capacity, specified by C_{COM}, and the admission control parameters N_{EL2}, N_{EL3}. These parameters are referred to as the output parameters of the system. The performance of the elastic POL policy can be tuned by the output parameters, and in particular, it has been realized that the setting of the output parameters C_{COM}, N_{EL2 }and N_{EL3}, allows for a tuning of the throughputtoblocking tradeoff for the elastic traffic classes.
With respect to the throughputtoblocking tradeoff for elastic traffic, the invention is generally directed towards the provision of a high link bandwidth utilization under one or more calllevel constraints that are related to at least one of elastic throughput and elastic blocking probability.
According to a preferred embodiment of the invention, the link capacity sharing algorithm aims at setting the output parameters of the elastic POL policy in such a way as to minimize call blocking probabilities B_{2 }and B_{3 }for elastic flows, while being able to take into account a blocking probability constraint (GoS) for the rigid flows as well as minimum throughput constraints for the elastic flows. The throughput constraints for the elastic flows are introduced because it has been recognized that there is a minimum acceptable throughput below which the users gain no actual positive utility.
Therefore, the rigid traffic class is associated with a maximum accepted call blocking probability B_{1} ^{max}, and the elastic adaptive and elastic nonadaptive traffic classes are associated with minimum accepted throughputs {tilde over (θ)}_{min }and {circumflex over (θ)}_{min}, respectively. Preferably, the throughput constraints are formed as constraints on the probability that the userperceived throughput during transfer of certain amount of data drops below a given threshold. Such a performance measure is easier for the user to verify than the traditional fairness criteria discussed in references [13].
Although the blocking probabilities for elastic traffic are being minimized, it is nevertheless normally advisable, although not necessary, to have a worstcase guarantee of the call blocking for elastic traffic, and associate also the two elastic traffic classes with maximum allowed block probabilities B_{2} ^{max }and B_{3} ^{max}.
In this case, the traffic input parameters of the system are the set of arrival rates (λ_{1}, λ_{2}, λ_{3}) and departure rates (μ_{1}, μ_{2}, μ_{3}) obtained from the network, the bandwidths (b_{1}, b_{2}, b_{3}), the minimum elastic bandwidth demands (b_{2} ^{min}, b_{3} ^{min}), the blocking probability constraints (B_{1} ^{max }or the whole set of B_{1} ^{max}, B_{2} ^{max }and B_{3} ^{max}) and the elastic throughput constraints ({tilde over (θ)}_{min }and {circumflex over (θ)}_{min}) The departure rate for nonadaptive class can be estimated under the assumption that the bandwidth of the nonadaptive flows equals b_{3}.
The parameters and performance measures associated with the rigid traffic class and the two elastic traffic classes are summarized in Table I below.
TABLE I  
Input parameters  System state  
Maximum  Performance  Number  
Call  Peak  Minimum  accepted  Minimum  measures  of flows  
Class  arrival  Departure  bandwidth  bandwidth  blocking  accepted  Through  in the  
rate  rate  requirement  requirement  probability  throughout  Blocking  put  system  
Rigid  λ_{1}  μ_{1}  b_{1}  —  B_{1} ^{max}  —  B_{1}  —  n_{1} 
Adaptive  λ_{2}  μ_{2}  b_{2}  b_{2} ^{min}  (B_{2} ^{max})  {tilde over (θ)}_{min}  B_{2}  {tilde over (θ)}  n_{2} 
elasic  
Non  λ_{3}  μ_{3}  b_{3}  b_{3} ^{min}  (B_{3} ^{max})  {circumflex over (θ)}_{min}  B_{3}  {circumflex over (θ)}  n_{3} 
adaptive  
elastic  
The problem of determining the output parameters of the elastic POL policy under blocking and throughput constraints is outlined below with reference to FIG. 2, which is a schematic flow diagram of the overall link capacity sharing algorithm according to a preferred embodiment of the invention. In the first step 101, the required input parameters, such as current arrival and departure rates, bandwidth requirements as well as the constraints imposed on the traffic, are provided. In step 102, the GoS (call blocking) requirement for rigid traffic is guaranteed by the proper setting of C_{COM}. In particular, we determine the minimum required capacity of C_{COM }for rigid flows that guarantees the reed blocking probability B_{1} ^{max}:
where B_{1 }is the blocking probability of rigid flows. For example, the wellknown ErlangB formula can be used to estimate such a value of C_{COM }based on arrival and departure rates and peak bandwidth requirement for the rigid traffic as inputs. In addition, a maximum number N_{COM }of admissible rigid flows can be determined based on the ErlangB analysis and used for admission control of the rigid traffic.
Next, we have to determine a maximum number of elastic flows (N_{EL2}, N_{EL3}) that can be simultaneously present in the system at the same time as the required throughput and blocking requirements are fulfilled. It is intuitively clear that if the maximum number N_{EL2 }of adaptive elastic flows is increased, the blocking probability B_{2 }of adaptive elastic flows decreases and the throughput decreases as well. Unfortunately, changing N_{EL2 }affects both the blocking probability B_{3 }and throughput of nonadaptive elastic flows and viceversa.
In this particular embodiment, the link capacity sharing algorithm a at minimizing the blocking probabilities of the elastic traffic classes under throughputthreshold constants. To accomplish this, the invention proposes an iterative procedure, generally defined by steps 103107, for tuning the cutoff parameters so that the throughputthreshold constraints are just fulfilled, no more and no less. First, in step 103, initial values of the cutoff parameters are estimated. Next, the performance of the system is analyzed (step 104) with respect to elastic throughputs. In particular, the throughputs {tilde over (θ)} and {circumflex over (θ)} offered in the system controlled by the initial values of the cutoff parameters are analyzed (step 104) and related (step 105) to the throughputthreshold constraints {tilde over (θ)}_{min }and {circumflex over (θ)}_{min}. If the offered throughputs are too low, then the cutoff parameters are reduced (step 106), increasing the blocking probabilities and also increasing the throughputs. On the other hand, if the offered throughputs are higher than the throughputthresholds, then the cutoff parameters can be increased (step 107) so that the blocking probabilities (as well as the throughputs) are reduced. In this way, by iteratively reputing the steps 104, 105 and 106/107, the blocking probabilities can be reduced to a minimum, while still adhering to the throughput constraints. Once the constraints are fulfilled to a satisfactory degree, the algorithm outputs (step 108) the parameters C_{COM}, (C_{ELA}), (N_{COM}), N_{EL2}, N_{EL3 }for use in controlling the actual bandwidth sharing of the considered transmission link.
Naturally, the steps 101 to 108 are repeated in response to changing traffic conditions so as to flexibly adapt the bandwidth sharing to the prevailing traffic situation.
In general, the cutoff parameters have to be reduced to fulfill the throughput constrain. On the other hand, as the aim is to minimize the elastic blocking probabilities, and as it is advisable to have a worstcase guarantee of the blocking probabilities for elastic traffic, the cutoff parameters must at the same time be as high as possible, and at least high enough to fulfill the worstcase blocking constraints. Depending on the model parameters and the given bounds, it may be the case that all the constraints can not be satisfied at the same time, which means that the link is overloaded with respect to the GoS requirements.
FIG. 3 is a schematic block diagram of pertinent parts of an IP router (or an ATM switch) in which a link capacity sharing algorithm according to the invention is implemented. The IP router 20 is associated with an input link and an output link. The router 20 has a control unit 21, a CAC unit 22, an output port buffer 23 for rigid traffic, an output port buffer 24 for elastic traffic, and an output port scheduler 25.
The control unit 21 is preferably, although not necessarily, realized as software on a computer system. The software may be written in almost any type of computer age, such as C, C++, Java or even specialized proprietary languages. In effect, the link capacity algorithm is mapped into a software program, which when executed on the computer system produces a set of output control parameters C_ELA, C_COM, N_ELA, N_COM in response to appropriate traffic input parameters received from the network and the UNCs by conventional means.
The N_ELA, N_COM parameters represents the cutoff parameters for rigid traffic and elastic traffic, respectively. In the example of FIG. 3, only a single elastic traffic class is considered, and hence only a single cutoff parameter N_ELA for elastic traffic is produced by the control unit 21. The cutoff parameters are forwarded to the CAC unit 22, which accepts or rejects new flows based on the forwarded cutoff parameters. For each requested new flow, the traffic class of the flow is determined so that admission control can be exercised based on the relevant cutoff parameter. IP packets belonging to accepted rigid flows (restricted by N_COM) are forwarded to the output port buffer 23 for subsequent scheduling by the output port scheduler 25. In the same way, IP packets belonging to accepted elastic flows (restricted by N_ELA) are forwarded to the output port buffer 24.
The C_ELA, C_COM parameters are forwarded from the control unit 21 to the output port scheduler 25. The output port scheduler 25 represents the bandwidth of the output link, and the actual bandwidth representation used in the traffic scheduling is determined by die C_ELA, C_COM parameters. In the output port scheduler 25, the bandwidth of the output link is divided into a common part C_COM, and a dedicated part C_ELA reserved for elastic traffic only. In scheduling IP packets, the output port scheduler 25 can use only the common bandwidth part C_COM for IP packets from the output port buffer 23 for rigid flows. For IP packets from the output port buffer 24 for elastic flows on the other hand, the scheduler 25 can use both the dedicated bandwidth part C_ELA and the common bandwidth part C_COM. In this way, the output port output port scheduler 25 decides how many IP packets that can be sent on the output link per time unit and traffic class.
Analysis of Throughput and Blocking Probability Measures of Elastic Flows
The throughput constraints used in the evaluation step 105 (FIG. 2) may for example be constraints on the average throughput, where the cutoff parameters fulfills the throughput constraints if:
where E stands for the expected value. To make a plausible interpretation of this type of constraints, let us assume that the distribution of θ is fairly symmetric around E(θ). In other words, the median of θ is close to E(θ). In this case, the probability that an elastic flow obtains less bandwidth than θ^{min }is around 0.5.
However, users often prefers more informative throughput constraints, and an alternative constraint may require that the throughput of adaptive and nonadaptive flows are greater than {tilde over (θ)}_{min }and {circumflex over (θ)}_{min }with predetermined probabilities (1−ε_{2}) and (1−ε_{3}), respectively, independent of the associated service requirements (x) or holding times (t):
The worstcase constraints on the elastic blocking probabilities can simply be expressed as:
In order to obtain the elastic throughput measures (step 104) and possibly also the elastic blocking me s for given values of the cutoff parameters so as to enable evaluation (step 105) against the given constraints, the steady state distribution of a Markov chain describing the dynamics of the mixed rigidelastic traffic needs to be determined. As implied in connection with the formulation of the multiclass model above, the system under investigation can be represented as a Continuous Time Markov Chain (CTMC), the state of which is uniquely characterized by the number of flows of the different traffic classes (n_{1}, n_{2}, n_{3}). It is clear that in order to obtain the perforce measures of the system we have to determine the CTMC's generator matrix Q and its steady state distribution P={P_{i}}, where P ^{T}·Q=0 and Σ_{i}P_{i}=1. The notions of a generator matrix and a steady state distribution of a Markov chain are considered well known to the skilled person. For a general introduction to loss networks, Markov theory and the general stochastic knapsack problem, reference is made to [8], and especially pages 169 thereof. For given values of the parameters C_{COM}, N_{EL2}, N_{EL3}, the set of triples (n_{1}, n_{2}, n_{3}) that satisfies the constraints of the elastic POL policy given by (14) constitute the set of feasible states of the system denoted by S. The cardinality of the state space can be determined as:
It is easy to realize that the generator matrix Q possesses a nice structure, because only transitions between “neighboring states” are allowed in the following sense. Let q_{ij }denote the transition rate from state i to state j. Then, taking into account the constraints (14) on the number of flows in the system defined by the elastic POL policy, the nonzero transition rates between the states are:
q _{i,i3−} =n _{3} ·r _{3}·μ_{3} (12)
where i_{1+}=(n_{1}+1, n_{2}, n_{3}) when i=(n_{1}, n_{2}, n_{3}); i_{k+} and i_{k−} (k=1, 2, 3) are defined similarly. Expression (10) represents the state transitions due to a call arrival, while (11) and (12) represent transitions due to call departures. The quantity defined in (12) denotes the total bandwidth of the nonadaptive flows when the system is in state i. The generator matrix Q of the CTMC is constructed based on the transition rates defined in (1012).
For illustrative purposes, let us consider a small system with a rigid class, an adaptive elastic class and a nonadaptive elastic class, where the link capacity C=7. For simplicity, assume a division of the link capacity such that n_{1}=1 is kept fixed, i.e. the available bandwidth for elastic flows is 6 bandwidth units. Furthermore, b_{1}=1, b_{2}=3 and b_{3}=2. The elastic flows are characterized by their minimum accepted bandwidths, which here are set to b_{2} ^{min}=1.8 and b_{3} ^{min}=0.8. Setting the cutoff parameters to N_{EL2}=2 and N_{EL3}=3, gives rise to 12 feasible states as illustrated in the Markov chain state space diagram of FIG. 4. There are 5 (gray) states where at least one of the elastic flows is compressed below the peak bandwidth specified by b_{2 }and b_{3}. The states are identified by the number of active connections (n_{1}, n_{2}, n_{3}). The values below the state identifiers indicate the bandwidth compression of the adaptive and nonadaptive elastic traffic (r_{2}, r_{3}). The state (1, 2, 3) is the only one where the bandwidth compression of the adaptive class and the nonadaptive class differs due to different minimum bandwidth requirements (r_{2} ^{min}=0.6, r_{3} ^{min}=0.4).
Different numerical solutions can be used to obtain the steady state distribution of a socalled multidimensional Markov chain. Direct methods such as the Gaussian elimination method compute the solution in a fixed number of operations. However, when considering the size of the state space for practically interesting cases, i.e. large state spaces in the order of 10^{4 }or higher, the computational complexity of the direct methods is usually unacceptable. Therefore, an iterative method, such as the biconjugate gradient method applied here, is much more feasible for the steady state analysis. The biconjugate gradient method is also detailed in reference [9].
The computation time of an iterative method depends on factors such as the speed of convergence and the complexity of each iteration step. The computation time is also highly dependent on the initial guess. A good initial guess will significantly reduce the overall computation time. For this reason, according to the invention, a heuristic direct method is applied for calculating a fairly close initial guess to be applied in the iterative method. Some special multidimensional Markov chains exhibit a so called product form solution, which means that the steady state probability of state (i,j) can be efficiently determined in product form as f(i)·g(j) instead of h(i,j). Unfortunately, due to the occasional reduction of the bandwidth (and corresponding departure rate) of the nonadaptive elastic flows, the CTMC of the studied system does not exhibit the nice properties of reversibility and product form solution, but the proposed initial guess used for the subsequent iterative numerical procedure is calculated as if the Markov chain exhibited product form. In other words, the initial form of the steady state distribution of a Markov chain describing a traffic system that includes nonadaptive elastic traffic is determined based on Markov chain product form calculations, and applied in an iterative steady state analysis method.
The fact that only nonadaptive elastic flows disturb the reversibility is utilized, and the Markov chain that describes the number of rigid and adaptive elastic flows in the system is reversible, and
is obtained from:
where the p*(n_{1},n_{2}) unnormalized steady state probabilities are auxiliary variables of the iterative method. From the steady state distribution of the rigid and adaptive flows (p(n_{1},n_{2})), the overall steady state behavior (p(n_{1},n_{2},n_{3})) is obtained by fixing the number of rigid flows (n_{i}=i) and assuming that the obtained Markov chain is reversible, even though this is not the case. This assumption allows us to evaluate an initial guess for the iterative method as follows. For all possible fixed values of n_{1 }(n_{1}=i):
In other words, we group states with common n_{1}, n_{2 }parameters, summing up their probabilities, to obtain a new 2dimensional Markov chain. The obtained 2dimensional Markov chain exhibits product form, and its steady state distribution is calculated using equations (1315). Next, we “share” the probability of the state groups among the individual states that define a state group using equations (1618).
The steady state distribution of traffic classes other than the nonadaptive traffic class is calculated as if there was no nonadaptive traffic in the system, and then state probabilities are calculated under the assumption of equilibrium between incoming and outgoing traffic of one of the other traffic classes and the nonadaptive elastic traffic class. In the present example, the steady state distribution of the rigid traffic class and the adaptive elastic traffic class is calculated as if there was no nonadaptive elastic traffic in the system, and state probabilities are determined assuming that the incoming and outgoing adaptive and nonadaptive elastic traffic are in equilibrium. It should though be understood that equations (1318) can be adapted to a variety of applications, for example traffic systems with several rigid traffic classes but only a single elastic traffic class.
It should also be understood that the above procedure for calculating an initial approximation of a steady state distribution is generally applicable to any multidimensional Markov chain and can be adapted to different applications,
The obtained initial approximation of the steady state distribution is used as a good initial guess for an iterative method, such as the biconjugate gradient based method, which improves the initial guess stepbystep to an appropriate accuracy.
Based on the steady state distribution of the CTMC, the call blocking probabilities can be calculated as:
The calculation of the average throughput of the adaptive and nonadaptive elastic flows is also quite straightforward once the steady state distribution of the CTMC is determined:
Thus, the blocking probability constraints in (5) and (8) as well as the average throughout constrains in (6) can be evaluated.
Unfortunately, it is much harder to check the throughput threshold constraints in (7), since neither the distribution nor the higher moments of {tilde over (θ)}_{t }and {circumflex over (θ)}_{x }can be analyzed based on the steady state distribution of the above studied Markov chain. Hence, a new analysis approach is applied. The throughput threshold constraint on adaptive elastic flows can be checked based on the distribution of {tilde over (θ)}_{t }and the throughput threshold constraint on nonadaptive elastic flows can be checked based on the distribution of T_{x}, because:
Since it is computationally to hard to evaluate the distribution of T_{x }and {tilde over (θ)}_{t }for realistic models, but there are effective numerical methods to obtain their moments, we check the throughput threshold constraint by applying a moment based distribution estimation method as disclosed in reference [10] and summarized in Table II below. In Table II, μ_{n }denotes the nth moment of the random variable X and the formulas present an upper and lower bound on the distribution of X. Table II is valid for any nonnegative random variable, i.e. we do not utilize the fact that T_{x }and {tilde over (θ)}_{t }are upper bounded in our system.
TABLE II  
Pr(X ≧ 1) ≦ upper limit  Pr(X ≧ 1) ≦ upper limit  
1  μ_{1}  0  
2 



3 



The method to evaluate the moments of T_{x }and {tilde over (θ)}_{t }is based on a tagging an elastic flow arriving to the system, and carefully exam the possible transitions from the instance this tagged flow enters the system until it leaves the system. The system behavior during the service of the tagged flow can be described by a slightly modified Markov chain. To analyze {tilde over (θ)}_{t }a tagged adaptive elastic flow is considered, while to analyze T_{x }a tagged nonadaptive elastic flow is considered. The modified system used to evaluate {tilde over (θ)}_{t }(or T_{x}) has the following properties:
Since it is assumed that at least the tagged elastic flow is present in the system we exclude states where n_{2}=0 (or n_{3}=0).
With each state of the state space there is an associated entrance probability, which is the probability of the event that the modified CTMC starts from that state. When the tagged elastic flow finds the system in state (n_{1},n_{2},n_{3}) it will bring the system into state (n_{1},n_{2}+1,n_{3}) (or state (n_{1},n_{2},n_{3}−1)) unless the state (n_{1},n_{2},n_{3}) happens to be a blocking state of the tagged flow.
Let {Z(t), t≧0} be the modified CTMC assuming the tagged elastic flow never leaves the system over the finite state space F with generator B. F can be defined as:
Indeed, F=S\S_{0 }where S_{0 }is the states in S where n_{2}=0 (or n_{3}=0). The state transition rates in B are closely related to the appropriate rates in Q.
The initial probability of the modified Markov chain p^{+}(n_{1},n_{2},n_{3}) is obtained by considering the system state immediately after the tagged flow joins the system in steady state. This means that the probability that the system is in state (n_{1},n_{2},n_{3}) after the tagged flow's arrival is proportional to the steady state probability of state (n_{1},n_{2}−1,n_{3}) (or (n_{1},n_{2},n_{3}−1)). Consequently:
To obtain the moments of {tilde over (θ)}_{t}, Markov Reward model is defined over {Z(t), t≧0} in accordance with reference [11]. {tilde over (θ)}_{t }is a random variable which depends on the random arrival and departure of the rigid, adaptive and nonadaptive elastic flows as described by B. The reward rate associated we the states of the modified Markov chain represents the bandwidth of the tagged adaptive elastic flows in that state. Let t_{i }be the reward rate (the bandwidth of the tagged adaptive elastic flow) in state i and T the diagonal matrix composed of the t_{i }entries. t_{1}=r_{2}(i)·b_{2}, where r_{2}(i) is the bandwidth compression in state i. In this way, the dynamics of the number of flows in the system during the service of the tagged flow is described by the Modified Markov chain, and the instantaneous bandwidth of the tagged flow is described by the instantaneous reward rate. If there are more flows in the system, the bandwidth of the tagged flow decreases towards b_{2} ^{min }and if there are less flows, it increases towards b_{2}. The generator matrix B and the reward matrix T define the Markov Reward model that accumulates t·{tilde over (θ)}_{t }amounts of reward in the interval (0, t). This means that the reward accumulated in the interval (0, t) represents the amount of data transmitted through the tagged flow in this interval, and {tilde over (θ)}_{t }is the amount of transmitted data/t.
T_{x }is the random amount of time that it takes to transmit x units of data through the tagged flow. By defining a Markov Reward model as above, the reward accumulated in the interval (0, t) represents the random amount of data transmitted through the tagged flow, and hence T_{x }is the time it takes to accumulate x amounts of reward. This measure is commonly referred to as completion time.
Having the initial probability distribution p^{2+}(n_{1},n_{2},n_{3}), and p^{3+}(n_{1},n_{2},n_{3}), the generator matrix B and the reward matrix T, the numerical analysis method proposed in reference [11] is applied to evaluate the moments of {tilde over (θ)}_{t }and T_{x}. This numerical method is applicable for Markov Reward models with large state spaces (^{−}10^{6 }states).
Numerical Examples of the Application of the Link Capacity Sharing Algorithm
By way of example, consider a transmission link of capacity C=100 Mbps and supporting tree different service classes: rigid, adaptive elastic and nonadaptive elastic service classes. The parameters, given as network traffic inputs and determined by the link sharing algorithm, of this system are as follows:
C_{COM}=20 Mbps, C_{ELA}=80 Mbps;
b_{1}=1 Mbps, b_{1}=5 Mbps, b_{3}=3 Mbps;
λ_{1}=λ_{3}=12 1/min;
μ_{1}=μ_{2}=μ_{3}=1 1/min;
r_{2} ^{min}=0.05, r_{3} ^{min}=0.001;
N_{COM}=20, N_{EL2}=20, N_{EL3}=20.
The effect of the arrival rate λ_{2 }of the adaptive elastic flows on the corresponding blocking probability for a number of values of minimum accepted throughput is demonstrated in Table III below.
TABLE III  
Pr({tilde over (θ)}_{t }≧ {tilde over (θ)}_{min}) ≧ (1ε_{2})  λ_{2 }= 12  λ_{2 }= 14  λ_{2 }= 16  
{tilde over (θ)}_{min }= 2.6  89.4%  83.6%  77.7%  
3  84.2%  75.1%  65.5%  
3.4  74.4%  62.0%  50.8%  
3.8  64.3%  47.8%  34.9%  
4.2  42.5%  23.4%  10.5%  
4.6  4.8%  0.14%  —  
As the minimum accepted throughput {tilde over (θ)}_{min }for the adaptive elastic traffic is assigned higher and higher values, the probability that an adaptive elastic flow obtains this throughput decreases. The increase of the arrival rate of the adaptive elastic flows results in more adaptive elastic flows in the system, and hence the throughput decreases together with the probability that the adaptive elastic flows obtain the required bandwidth.
The effect of the arrival rate λ_{3 }of the nonadaptive elastic flows on the corresponding blocking probability for a number of values of minimum accepted throughput is demonstrated in Table IV below. In this case, the system parameters are:
C=250 Mbps;
C_{COM}=50 Mbps, C_{ELA}=200 Mbps;
b_{1}=1 Mbps, b_{2}=3 Mbps, b_{3}=5 Mbps;
λ_{1}=40 1/min, λ=25 1/min;
μ_{1}=μ_{2}=μ_{3}=1 1/min;
r_{2} ^{min}=0.4, r_{3} ^{min}=0.05;
N_{COM}=50, N_{EL2}=120, N_{EL3}=180.
Note that in this case the modified Markov chain describing the system behavior during the service of a tagged nonadaptive elastic flow has 1,116,951 states and 6,627,100 transitions.
TABLE IV  
Pr({tilde over (θ)}_{x }≧ {circumflex over (θ)}_{min}) ≧ (1ε_{2})  λ_{3 }= 20  λ_{3 }= 25  λ_{3 }= 30  
{circumflex over (θ)}_{min }= 2.5  99.98%  99.6%  88.1%  
3.33  99.8%  94.36%  32.5%  
4.0  97.4%  68.1%  13.8%  
4.34  91.5%  59.8%  —  
4.54  89.6%  52.3%  —  
4.76  86.0%  30.9%  —  
In similarity to the effects demonstrated in Table III, as the minimum accepted throughput for the nonadaptive elastic traffic is assigned higher and higher values, the probability that a nonadaptive elastic flow obtains this throughput decreases. Also, the increase of the arrival rate results in a decreasing probability that the nonadaptive elastic flows obtain the required bandwidth.
To get an impression of the relation of average throughput and throughput threshold constraints reference is made to FIG. 5, which illustrates the mean and the variance of the throughput of adaptive elastic flows as a function of their service time. The graph of FIG. 5 relates to the system considered for Table III, with λ_{2}=14. The mean throughput is shown by a solid line, whereas the variance is shown by a dashed line. It can thus be seen that for “short” (with respect to service time) connections, the variance of the throughput is quite significant, and consequently, the average throughput and the throughput threshold constraints have significantly different meaning. For “long” connections, the variance of the throughput almost vanishes, and the mean throughput provides a meaningful description of the bandwidth available for adaptive elastic flows. Note that {tilde over (θ)}_{t }tends to approach a deterministic value, the steady state throughput, as t goes to infinity.
Finally, we study an example of how to select N_{EL2 }and N_{EL3 }to provide the required QoS parameters. Assume that after the division of the link capacity and the dimensioning of the rigid class, the systems parameters have the following values:
C=100 Mbps;
C_{COM}=20 Mbps, C_{ELA}=80 Mbps;
b_{1}=1 Mbps, b_{2}=5 Mbps, b_{3}=3 Mbps;
λ_{1}=12 1/min, λ_{2}=12 1/min, λ_{3}=12 1/min;
μ_{1}=μ_{2}=μ_{3}=1 s (here expressed as mean holding time);
b_{3} ^{min}=0.1 Mbps;
The parameters N_{EL2 }and N_{EL3 }have to be such that the elastic blocking probabilities are less than 1% (B_{2}<0.01, B_{3}<0.01) and the average throughput parameters fulfill E({tilde over (θ)})≧4.05 and E({circumflex over (θ)})≧2.35.
The set of N_{EL2 }and N_{EL3 }parameters that fulfill the QoS requirements are depicted in the gray area of FIG. 6. The blocking probability limit of the adaptive elastic class is a vertical line due to the independence on the load of the nonadaptive elastic class. The blocking probability limit of the nonadaptive elastic class is a horizontal line. With the considered low level of overall load, the average elastic throughputs are hardly sensitive to the N_{EL2 }and N_{EL3 }parameters after a given limit. In this example, the tighter of the two bandwidth limits that determines the acceptable N_{EL2 }and N_{EL3 }values, is the E({tilde over (θ)})≧4.05 bound.
Inversion of the Optimization Task
The new elastic POL policy allows for a natural inversion of the optimization task so that instead of minimizing blocking probabilities for elastic traffic under throughput constraints, the elastic throughputs are maximized under blocking probability constraints. In similarity to the link capacity sharing method illustrated in the flow diagram of FIG. 2, traffic input parameters are received (similar to step 101), the common link capacity part C_{COM }is determined (similar to step 102) and initial values of the cutoff parameters are selected (similar to step 103). Next, the performance of the system is analyzed (similar to step 104), but now primarily with respect to elastic blocking probabilities. In particular, the elastic blocking probabilities in the system are analyzed and related (similar to step 105) to the blocking probability constraints. If the blocking probabilities are too high, then the cutoff parameters are increased, reducing the blocking probabilities and also reducing the throughputs. On the other hand, if the blocking probabilities are lower than the blocking constraints, then the cutoff parameters can be reduced so that the blocking probabilities as well as the throughputs are increased. In this way, by way of iteration, the throughputs can be increased to a maximum, while still adhering to the blocking constraints for elastic flows. As the aim now is to maximize the elastic throughputs, and as it might be advisable to have a worstcase guarantee for the through of elastic traffic, the cutoff parameters must be as low as possible, and at least low enough to fulfill the worstcase throughput constraints, while still fulfill the blocking probability constraints imposed on the elastic traffic.
Naturally, the link capacity sharing algorithm, irrespective of whether it is adapted for minimizing elastic blocking or for maximizing elastic throughput, is also applicable to elastic traffic only. For example, in the absence of rigid traffic, C_{COM }is reduced to zero, and the overall link capacity sharing algorithm is reduced to the mathematical formulas for determining the cutoff parameters under throughput/blocking constraints. Furthermore, in the case of a single elastic traffic class, only a single cutoff parameter needs to be determined according to the above iterative link sharing algorithm.
Although, the link capacity sharing algorithm has been described above with reference to an IP network carrying a single rigid traffic class and two different elastic traffic classes, it should be understood that the invention is not limited thereto, and that the algorithm is applicable to other types of networks and other traffic classes. In fact, an example of the elastic POL algorithm applied in an ATM network carrying narrowband CBR (Constant Bit Rate) traffic and wideband CBR traffic, as well as ABR (Available Bit Rate) traffic will be outlined below.
In this example, calls arriving at a transmission link generally belong to one of the following three traffic classes:
Class 1—Narrowband CBR calls, characterized by their peak bandwidth requirement b_{1}, call arrival rate λ_{1 }and departure rate μ_{1}.
Class 2—Wideband CBR calls, characterized by their peak bandwidth requirement b_{2}, call arrival rate λ_{2 }and departure rate μ_{2}.
Class 3—ABR calls, characterized by their peak bandwidth requirement b_{3}, minimum bandwidth requirement b_{3} ^{min}, call arrival rate λ_{3 }and ideal departure rate μ_{3}. The ideal departure rate is experienced when the peak bandwidth is available during the entire duration of the call.
It should be noted that the CBR classes can be likened by the rigid traffic class of the above IP network example, and that the ABR class can be likened by the nonadaptive elastic traffic class described above in connection with the IP network example. In this respect, the assumptions in the model formulated in the IP network example are equally applicable in the present example.
The elastic POL policy described above is applied to the mixed CBRABR traffic environment in the ATM network considered. This means that the link capacity C is divided into two parts, a common part C_{COM }for CBR calls as well as ABR calls, and a dedicated part C_{ABR}, which is reserved for the ABR calls, such that C=C_{COM}+C_{ABR}. An admission control parameter N_{ABR}, also referred to as a cutoff parameter, is introduced for the ABR calls. Under the elastic POL policy, the number n_{1}, n_{2 }and n_{3 }of narrowband CBR, wideband CBR and ABR calls, respectively, in progress on the link is subject to the following constraints:
In (1) the ABR calls are protected from CBR calls. In (23) the maximum number of ABR calls is limited by two constraints. Expression (2) protects CBR calls form ABR calls, while (3) protects the inprogress ABR calls from new ABR calls. In this case, the elastic POL policy is fully determined by the division of the lint capacity, specified by C_{COM}, and the admission control parameter N_{ABR}. The performance of the elastic POL policy is tuned by these parameters.
According to a preferred embodiment of the invention, the link capacity sharing algorithm aims at setting the output parameters C_{COM }and N_{ABR }of the elastic POL policy in such a way as to minimize the call blocking probability for the ABR calls, while being able to take into account blocking probability constraints (GoS) for the different types of CBR calls and a minimum throughput constraint for the ABR calls. Therefore, each CBR class is associated with a maximum accepted call blocking probability B_{1} ^{max }and B_{2} ^{max}, and the ABR class is associated with a minimum accepted throughput θ_{min}, which can be treated in similarity to the minimum accepted throughput {circumflex over (θ)}_{min }for the nonadaptive elastic traffic of the IP network example.
Although the ABR blocking probability is being minimized, it is nevertheless normally advisable, although not necessary, to have a worstcase guarantee of the call blocking probability for ABR calls, and associate also the ABR class with a maximum allowed blocking probability B_{3} ^{max}.
The parameters and performance measures associated with the CBR classes and the ABR class are summarized in Table V below.
TABLE V  
Input parameters  System state  
Maximum  Performance  Number  
Call  Peak  Minimum  accepted  Minimum  measures  of flows  
Class  arrival  Departure  bandwidth  bandwidth  blocking  accepted  Through  in the  
rate  rate  requirement  requirement  probability  throughout  Blocking  put  system  
NCBR  λ_{1}  μ_{1}  b_{1}  —  B_{1} ^{max}  —  B_{1}  —  n_{1} 
WCBR  λ_{2}  μ_{2}  b_{2}  —  B_{2} ^{max}  B_{2}  —  n_{2}  
ABR  λ_{3}  μ_{3}  b_{3}  b_{3} ^{min}  (B_{3} ^{max})  θ_{min}  B_{3}  θ  n_{3} 
The problem of determining the output parameters of the elastic POL policy under the above constraints is outlined below with reference to FIG. 7, which is a schematic flow diagram of the overall link capacity sharing algorithm for a mixed CBRABR traffic environment according to a preferred embodiment of the invention. In the first step 201, the required input parameters are provided. In step 202, the GoS (call blocking) requirement for CBR traffic is guaranteed by the proper setting of C_{COM}. In particular, we determine the minimum required capacity of C_{COM }for CBR calls that guarantees the required blocking probabilities B_{1} ^{max }and B_{2} ^{max}:
For example, the wellknown ErlangB formula can be used to estimate such a value of C_{COM }based on arrival and de e rates and peak bandwidth requirements for the CBR classes as inputs.
Next, we have to determine a maximum number of ABR calls (N_{ABR}) that can be simultaneously present in the system at the same time as the required throughput and blocking requirements are fulfilled.
In this particular embodiment, the link capacity sharing algorithm aims at minimizing the blocking probability of the ABR calls under a minimum throughput constraint. To accomplish this, the invention proposes an iterative procedure, generally defined by steps 203207, for tuning the cutoff parameter so that the throughputthreshold constraint is just fulfilled, generally no more and no less. First, in step 203, an initial value of the cutoff parameter is estimated. Next, the performance of the system is analyzed (step 204) with respect to the ABR throughput, and related (step 205) to the throughputthreshold constraint. If the ABR throughput is too low, then the cutoff parameter is reduced (step 206), increasing the blocking probability and also increasing the throughput. On the other hand, if the ABR throughput is higher than the throughputthreshold, then the cutoff parameter can be increased (step 207) so that the blocking probability (as well as the throughput) is reduced. In this way, by iteratively repeating the steps 204, 205 and 206/207, the ABR blocking probability can be reduced to a minimum, while still adhering to the throughput constraint.
Preferably, the performance measures, ABR throughput and possibly also ABR blocking, are analyzed in more or less the same way as described above in connection with the IP network example. In short, this means determining the steady state distribution of the Markov chain that describes the dynamics and behavior of the mixed CBRABR environment, and calculating blocking and throughput measures based on the determined distribution. It should though be noted that here the throughputthreshold constraint, analogous to expression (7), is checked based on the transient analysis of the Markov chain that describes the mixed CBRABR environment using the numerical method proposed in reference [11] and applying the Markov inequality.
It is of course possible to invert the optimization task also for the ATM network example, in substantially the same manner as explained above for the IP network example.
Numerical Examples of the Application of the Link Capacity Sharing Algorithm
By way of example, consider an ATM transmission link of capacity C=155 Mbps and supporting three different service classes; two CBR classes and an ABR class, as described above. The input parameters of this ATM transmission link system are:
b_{1 }(nCBR)=3 Mbps, b_{2 }(wCBR)=6 Mbps, b_{3 }(ABR)=10 Mb;ps
λ_{1}=6 1/s, λ_{2}=3 1/s, λ_{3}=12 1/s;
μ_{1}=μ_{2}=μ_{3}=1 1/min;
r_{2} ^{min}=0.05, r_{3} ^{min}=0.001;
N_{COM}=50.
Furthermore, it is required that the blocking probabilities of the narrowband and wideband CBR calls are less than B_{1} ^{max}=2% and B_{2} ^{max}=4%, respectively. It thus follows that the minimal bandwidth for C_{COM }necessary to provide these blocking probabilities is 60 Mbps, which leaves C_{ABR}=95 Mbps for the ABR calls.
To examine the tradeoff between throughput and blocking probability for the ABR traffic, reference is made to Table VI below, which illustrates the average throughput E(θ) and the blocking probability B_{3 }for the ABR traffic class for different values of N_{ABR}.
TABLE VI  
N_{ABR}  10  20  40  60  80  100  150 
B_{3}  0.310  0.0811  0.0320  0.0212  0.00141  0.00112  0.000461 
E(θ)  9.99  7.9  4.83  3.45  2.69  2.2  1.52 
From Table VI, the tradeoff between throughput and blocking is apparent; highblocking=high throughput, and low blocking=low throughput. In the elastic POL policy according to the invention, this tradeoff is conveniently controlled by means of the N_{ABR }cutoff parameter as can be seen from Table VI. For instance, when constraint θ^{min }on the average throughput is set to 2.2 Mbps, the maximum number (N_{ABR}) of simultaneously active ABR calls is limited to 100.
In simulations, it has been observed that the elastic POL policy is superior to the wellknown Complete Partitioning (CP) policy under all loads, both in terms of blocking probabilities and ABR throughput. This is partly due to the fact that the POL policy allows ABR calls to make use of any bandwidth of the C_{COM }part currently not used by CBR calls.
Finally, to examine the impact of the C_{COM }parameter on the blocking probabilities and the average throughput of ABR traffic, reference is made to Table VII below.
TABLE VII  
C_{COM}  69  66  63  60  57  54 
B_{1}  0.00498  0.00770  0.0116  0.0171  0.0244  0.0342 
B_{2}  0.0126  0.0192  0.0284  0.0411  0.0578  0.0794 
B_{3}  0.0149  0.0141  0.0129  0.0115  0.00973  0.00773 
E(θ)  2.04  2.08  2.13  2.20  2.31  2.46 
The C_{COM }parameter offers a way of controlling the tradeoff between the CBR blocking probabilities on one hand, and the ABR blocking probability and throughput on the other hand. Prom Table VII, it can be seen that both the ABR throughput (increasing) and the ABR blocking (decreasing) are improved at the expense of degrading CBR blocking probabilities.
It is important to understand that the preceding description is intended to serve as a framework for an understanding of the invention. The embodiments described above are merely given as examples, and it should be understood that the present invention is not limited thereto. Further modifications, changes and improvements which retain the basic underlying principles disclosed and claimed herein are within the scope and spirit of the invention.
[1] L. Massoulie, J. Roberts, “Bandwidth Sharing: Objectives and Algorithms”, IEEE Infocom '99, pp. 13951403, March 1999.
[2] L. Massoulie, J. Roberts, “Bandwidth Sharing and Admission Control for Elastic Traffic”, ITC Specialist Seminar, Yokohama, October 1998.
[3] L. Massoulie, J. Roberts, “Arguments in Favour of Admission Control for TCP Plows”, 16^{th }International Teletraffic Congress, Edinburgh, UK, June, 1999.
[4] R. J. Gibbens and F. P. Kelly, “Distributed Connection Acceptance Control for a Connectionless Network”, 16^{th }International Teletraffic Congress, Edinburgh, UK, June, 1999.
[5] F. P. Kelly, “Charging and Rate Control for Elastic Traffic”, European Transaction on Telecommunications, pp. 3337, Vol. 8, 1997.
[6] Wuchang Feng, Dilip D. Kandlur, Debanjan Saha and Kang. G. Shin, “Understanding and Improving TCP Performance Over Networks with Minimum Rate Guarantees”, IEEE/ACM Transactions on Networking, pp. 173187, Vol. 7, No. 2, April 1999.
[7] E. D. Sykas, K. M. Vlakos, I. S. Venieris, E. N. Protonotarios, “Simulative Analysis of Optimal Resource Allocation and Routing in IBCN's”, IEEE JSAC, Vol. 9, No. 3, 1991.
[8] Keith W. Ross, “Multiservice Loss Models for Broadband Telecommunication Networks”, SpringerVerlag, 1995, ISBN 3540199187.
[9] W. J. Stewart, “Introduction to the Numerical Solution of Markov Chains”, pp. 220221, Princeton University Press, Princeton, N.J., ISBN 0691036993, 1994.
[10] M. Frontini, A. Tagliani, “Entropyconvergence in Stieltjes and Hamburger moment problem”, Appl. Math. and Comp., 88, pp. 3951, 1997.
[11] M. Telek and S. Rácz, “Numerical analysis of large Markov reward models”, Performance Evaluation, 36&37:95114, August 1999.
[12] A. Smith, J. Adams, G. Tagg, “Available Bit Rate—A New Service for ATM”, Computer Networks and ISDN Systems, 28, pp. 635640, 1996.
Cited Patent  Filing date  Publication date  Applicant  Title 

US5274644  Nov 5, 1991  Dec 28, 1993  At&T Bell Laboratories  Efficient, ratebase multiclass access control 
US5583792 *  May 27, 1994  Dec 10, 1996  SanQi Li  Method and apparatus for integration of traffic measurement and queueing performance evaluation in a network system 
US5881049  Oct 3, 1996  Mar 9, 1999  Northern Telecom Limited  Admission control in an ATM switching node 
US5909443  Jan 3, 1997  Jun 1, 1999  International Business Machines Corporation  ATM network congestion control system using explicit rate cell marking 
US5914945  Dec 31, 1996  Jun 22, 1999  Northern Telecom Limited  Method and system for bandwidth allocation for multimedia services under aggregate traffic conditions 
US6072800 *  Aug 18, 1997  Jun 6, 2000  Nec Usa, Inc.  Weighted longest queue first adaptive scheduling discipline for ATM networks 
US6115359 *  Dec 30, 1997  Sep 5, 2000  Nortel Networks Corporation  Elastic bandwidth explicit rate (ER) ABR flow control for ATM switches 
US6118764 *  Dec 30, 1997  Sep 12, 2000  Nortel Networks Corporation  Congestion indication/no increase (CI/NI) ABR flow control for ATM switches 
US6266322 *  Jun 5, 1998  Jul 24, 2001  At&T Corp.  Dimensioning bandwidth and connection admission control for elastic traffic in highspeed communication networks 
US6366559 *  Jun 5, 1998  Apr 2, 2002  Telcordia Technologies, Inc.  Method and system for statedependent admission control and routing of multirate circuitswitched traffic 
US6418139 *  Nov 25, 1998  Jul 9, 2002  Nortel Networks Limited  Mechanism to guarantee quality of service to realtime traffic on IP networks 
WO1997001895A2  Jun 27, 1996  Jan 16, 1997  Newbridge Networks Corporation  Connection admission control system (cac) for atm networks 
WO1998028938A1  Dec 3, 1997  Jul 2, 1998  Northern Telecom Limited  Dynamic traffic conditioning 
WO1998041052A1  Mar 4, 1998  Sep 17, 1998  Nokia Telecommunications Oy  Connection admission control in broadband network 
WO1999011003A1  Aug 27, 1998  Mar 4, 1999  Extreme Networks  Policy based quality of service 
WO1999034544A1  Dec 23, 1998  Jul 8, 1999  Ukiah Software, Inc.  Traffic monitoring tool for bandwidth management 
Reference  

1  "Traffic Control and Congestion Control in BISDN," ITUT Recommendation I.371, Traffic Management Specification (Helsinki, Mar. 1993), pp. 126.  
2  Bobbio, A., et al.; "Computation of the Distribution of the Completion Time When the Work Requirement is a PH Random Variable," Commun. Statist.Stochastic Models, 6(1), 1990, pp. 133150.  
3  Bonomi, F., et al.; "The RateBased Flow Control Framework for the Available Bit Rate ATM Service," IEEE Network, Mar./Apr. 1995, pp. 2539.  
4  Borst, S., Virtual Partitioning for Robust Resource Sharing: Computational Techniques for Heterogeneous Traffic, IEEE Journal on Selected Areas in Communications, vol. 15, No. 5, Jun. 1998, pp. 668678.  
5  Chen, T., et al., "The Available Bit Rate Service for Data in ATM Networks," IEEE Communications Magazine, May 1996, pp. 5671.  
6  Choudhury, G., et al., "Efficiently Providing Multiple Grades of Service with Protection Against Overloads in Shared Resources," AT&T Technical Journal, Jul./Aug. 1995, pp. 5063.  
7  DeSerres, Y., et al., "A Multiserver Queue with Narrow and WideBand Customers and WideBand Restricted Access," IEEE Transactions on Communications, vol. 36, No. 6, Jun. 1988, pp. 675684.  
8  Dziong, Z.,et al., "Call Admission and Routing in MultiService Loss Networks," IEEE Transactions on Communication, vol. 42, No. 2/3/4, Feb./Mar./Apr. 1994, pp. 20112022.  
9  Faragó, A., et al., "A New Degree of Freedom in ATM Network Dimensioning: Optimizing the Logical Configuration," IEEE Journal of Selected Areas in Communications, vol., 13, No., 7, Sep. 1995, pp. 11991206.  
10  Feng, W., et al.; "Understanding and Improving TCP Performance Over Networks with Minimum Rate Guarantees," IEEE/ACM Transactions on Networking, vol. 7, No. 2, Apr. 1999, pp. 173187.  
11  Fodor, G. et al., "Simulative Analysis of Routing and Link Allocation Strategies in ATM Networks Supporting ABR Services," IEICE Trans. Commun.vol. E81B No. 5, May 1998, pp. 985995.  
12  Fodor, G., et al., "Revenue Optimization and Fairness Control of Priced Guaranteed and Best Effort Services on an ATM Transmission Link," IEEE International Conference on Communications, ICC '98, Atlanta, GA, Jun. 1998, pp. 16961705.  
13  Frontini, M., et al; "EntropyConvergence in Stieltjes and Hamburger Moment Problem," Applied Mathematics and Computation, 88:3951 (1997); Elsevier Science, Inc., New York, NY, pp. 3951.  
14  Gibbens, R.J., et al.; "Distributed Connection Acceptance Control for a Connectionless Network,"ITC 16/16<th >International Teletraffic Congress, Edinburgh, UK, Jun. 1999 pp. 941952.  
15  Gibbens, R.J., et al.; "Distributed Connection Acceptance Control for a Connectionless Network,"ITC 16/16th International Teletraffic Congress, Edinburgh, UK, Jun. 1999 pp. 941952.  
16  International Search Report completed by the ISA/SE on Mar. 20, 2001 in connection with priority application PCT/SE00/01827 as mailed on Mar. 23, 2001.  
17  Kaufman, J., "Blocking in a Shared Resource Environment," IEEE Transactions on Communications, vol. Com29, No. 10, Oct. 1981, pp. 14741481.  
18  Kelly, F., "Charging and Rate Control for Elastic Traffic," European Transactions on Telecommunications, vol. 8 (1997), pp. 111 [corrected version].  
19  Massoulié, L., et al.; "Bandwidth Sharing: Objectives and Algorithms," France TelecomCNET, IEEE Jun. 1999, pp. 13951403.  
20  Massoulié, L., et al; "Arguments in Favour of Admission Control for TCP Flows," 16<th >International Teletraffic Congress, Edinburgh, UK, Jun. 1999, pp. 116.  
21  Massoulié, L., et al; "Arguments in Favour of Admission Control for TCP Flows," 16th International Teletraffic Congress, Edinburgh, UK, Jun. 1999, pp. 116.  
22  Mitra, D., et al., "ATM Network Design and Optimization: A Multirate Loss Network Framework," IEEE/ACM Transactions on Networking, vol. 4. No. 4, Aug. 1996, pp. 531543.  
23  Mitra, D., et al., "Robust Dynamic Admission Control for Unified Cell and Call QoS in Statistical Multiplexers," IEEE Journal on Selected Areas in Communications, vol. 16, No. 5, Jun. 1998, pp. 692707.  
24  Nilsson, A.A., et al, "Multirate Blocking Probabilities: Numerically Stable Computations," ITC 15 / International Teletraffic Congress, Washington, D.C., Jun. 1997, pp. 13591368.  
25  Nordström, E., "NearOptimal Link Allocation of Blockable NarrowBand and Queueable WideBand Call Traffic in ATM Networks," ITC 15 / International Teletraffic Congress, Washington, D.C., Jun. 1997, pp. 987996.  
26  Roberts, J.W., "Connection Admission Control," Methods for the Performance Evaluation and Design of Broadband Multiservice Networks, Published by the Commission of European Communities, Information Technology and Sciences, COST 242, Final Report, 1996, pp. 115171.  
27  Roberts, J.W., "Quality of Service Guarantees and Charging in Multiservice Networks," IEICE Transactions on Communications, Spec. Issue on ATM Traffic Control and Performance Evaluation, vol. E81B, No. 5, May 1998, pp. 824831.  
28  Roberts, J.W., "Realizing Quality of Service Guarantees in Multiservice Networks," Performance and Management of Complex Communication Networks, T. Hasegawa, et al (Eds.) 1998, pp. 277293.  
29  Roberts, J.W., et al.; "Bandwidth Sharing and Admission Control for Elastic Traffic," France TelecommCNET, ITC Seminar, Yokohama (1998), pp. 19.  
30  Rosenberg, S., et al., "Functionality at the Edge: Designing Scalable Multiservie ATM Networks," IEEE Communications Magazine, May 1998, pp. 8899.  
31  Ross, K.W., et al, "Optimal Circuit Access Policies in an ISDN Environment: A Markov Decision Approach," IEEE Transactions on Communications, vol. 37, No. 9, Sep. 1989, pp. 934939.  
32  Ross, K.W., MultiService Loss Models for Broadband Telecommunications Networks, Springer Verlag London Limited, ISBN 3540199187 (1995) "Chapter 2: The Stochastic Knapsack," pp. 1770.  
33  Ross, K.W., MultiService Loss Models for Broadband Telecommunications Networks, Springer Verlag London Limited, ISBN 3540199187 (1995), "Multiservice Loss Systems, Chapter 1: 1.11.6; The Stochastic Knapsack," Chapter 2: 2.12.10, pp. 168.  
34  Smith, A., et al., "Available Bit RateA New Service for ATM," Computer Networks and ISDN Systems, 28, (1996) pp. 635640.  
35  Smith, A., et al., "Available Bit Rate—A New Service for ATM," Computer Networks and ISDN Systems, 28, (1996) pp. 635640.  
36  Sykos, E.D., et al., "Simulative Analysis of Optimal Resource Allocation and Routing in IBCN's," IEEE Journal on Selected Areas in Communications, vol. 9, No. 3, Apr. 1991, pp. 486492.  
37  Telek, M., et al, "Numerical Analysis of Large Markov Reward Models," Performance Evaluation 3637 (1999), pp. 95114. 
Citing Patent  Filing date  Publication date  Applicant  Title 

US6931251 *  Dec 5, 2002  Aug 16, 2005  Motorola, Inc.  System and method of call admission control in a wireless network 
US6967921 *  Nov 27, 2000  Nov 22, 2005  At&T Corp.  Method and device for efficient bandwidth management 
US7065048 *  Jul 6, 2004  Jun 20, 2006  At&T Corp.  Method and device for efficient bandwidth management 
US7110365 *  Nov 28, 2001  Sep 19, 2006  Samsung Electronics Co., Ltd.  Apparatus for analyzing performance of traffic in asynchronous transfer mode (ATM) switch and method thereof, and ATM switching system employing the same 
US7136356 *  Mar 16, 2001  Nov 14, 2006  Hitachi, Ltd.  Packet data transfer method and packet data transfer apparatus 
US7363371 *  Dec 28, 2000  Apr 22, 2008  Nortel Networks Limited  Traffic flow management in a communications network 
US7492711 *  Aug 21, 2003  Feb 17, 2009  Cisco Technology, Inc.  Link sizing based on both user behavior and traffic characteristics 
US7630317 *  Dec 8, 2009  Fujitsu Limited  Transmission bandwidth control device  
US7782773 *  Aug 24, 2010  Cariden Technologies, Inc.  Metric optimization for traffic engineering in a metricrouted network  
US7826349 *  May 30, 2006  Nov 2, 2010  Intel Corporation  Connection management mechanism 
US7907539 *  Nov 7, 2007  Mar 15, 2011  At&T Mobility Ii Llc  Systems and methods for calculating call blocking for alternate call routing schemes 
US7929434 *  Jan 14, 2004  Apr 19, 2011  Siemens Aktiengesellschaft  Method for determining limits for controlling traffic in communication networks with access control 
US8027255  Sep 30, 2008  Sep 27, 2011  Alcatel Lucent  Method and apparatus for prioritizing packets for use in managing packets in radio access networks 
US8341288 *  Oct 22, 2004  Dec 25, 2012  Cisco Technology, Inc.  Mechanism for sharing resources among different senders and receivers 
US8369220 *  Feb 5, 2013  Avaya Inc.  Routing a flow of elastic traffic  
US8400926 *  Nov 30, 2007  Mar 19, 2013  Patentmarks Portfolio, Llc  Multiprotocol telecommunications routing optimization 
US8503432  Sep 30, 2008  Aug 6, 2013  Alcatel Lucent  Method and apparatus for signaling proprietary information between network elements of a core network in a wireless communication network 
US8553554  May 16, 2008  Oct 8, 2013  Alcatel Lucent  Method and apparatus for providing congestion control in radio access networks 
US8595374 *  Dec 8, 2010  Nov 26, 2013  At&T Intellectual Property I, L.P.  Method and apparatus for capacity dimensioning in a communication network 
US8595478  Nov 19, 2007  Nov 26, 2013  AlterWAN Inc.  Wide area network with high quality of service 
US8867529  Sep 20, 2010  Oct 21, 2014  Cisco Technology, Inc.  System and method for providing a fate sharing identifier in a network environment 
US9015471  Sep 10, 2013  Apr 21, 2015  Alterwan, Inc.  Interautonomous networking involving multiple service providers 
US9032450 *  Jun 30, 2011  May 12, 2015  Samsung Electronics Co., Ltd.  Adaptive audio/video streams proxy 
US9036499 *  Mar 13, 2013  May 19, 2015  Patentmarks Communications, Llc  Multiprotocol telecommunications routing optimization 
US9232027 *  Aug 25, 2008  Jan 5, 2016  International Business Machines Corporation  TCP connection resource diversity using tunable geometric series 
US9258240 *  Oct 19, 2010  Feb 9, 2016  Nec Corporation  Available bandwidth estimating device 
US9270725  Nov 25, 2013  Feb 23, 2016  At&T Intellectual Property I, L.P.  Method and apparatus for capacity dimensioning in a communication network 
US9369485 *  Dec 11, 2015  Jun 14, 2016  International Business Machines Corporation  TCP connection resource diversity using tunable geometric series 
US9414401 *  Dec 15, 2008  Aug 9, 2016  At&T Intellectual Property I, L.P.  Opportunistic service management for elastic applications 
US20010048662 *  Mar 16, 2001  Dec 6, 2001  Hitachi, Ltd.  Packet data transfer method and packet data transfer apparatus 
US20020039349 *  Apr 27, 2001  Apr 4, 2002  Malaney Robert Anderson  Telecommunications traffic regulator 
US20020131373 *  Nov 28, 2001  Sep 19, 2002  Samsung Electronics Co., Ltd.  Apparatus for analyzing performance of traffic in asynchronous transfer mode (ATM) switch and method thereof, and ATM switching system employing the same 
US20030152070 *  Feb 4, 2003  Aug 14, 2003  Siemens Aktiengesellschaft  Method for transmitting signaling messages between first and second network units, and radio communication system and base station subsystem therefor 
US20040042398 *  Feb 27, 2003  Mar 4, 2004  Seriqa Networks  Method and apparatus for reducing traffic congestion by preventing allocation of the occupied portion of the link capacity and for protecting a switch from congestion by preventing allocation on some of its links 
US20040110507 *  Dec 5, 2002  Jun 10, 2004  Ramakrishnan Kajamalai J  System and method of call admission control in a wireless network 
US20040184483 *  Jan 30, 2004  Sep 23, 2004  Akiko Okamura  Transmission bandwidth control device 
US20060056299 *  Jan 14, 2004  Mar 16, 2006  Michael Menth  Method for determining limits for controlling traffic in communication networks with access control 
US20060089988 *  Oct 22, 2004  Apr 27, 2006  Davie Bruce S  Mechanism for sharing resources among different senders and receivers 
US20080005314 *  May 30, 2006  Jan 3, 2008  Sumeet Kaur  Connection management mechanism 
US20080019381 *  Jul 21, 2006  Jan 24, 2008  Mills David W  System And Method For Establishing A Communication Session Between Two Endpoints That Do Not Both Support Secure Media 
US20080123534 *  Jan 31, 2008  May 29, 2008  Cariden Technologies, Inc.  Metric optimization for traffic engineering in a metricrouted network 
US20080225832 *  Nov 30, 2007  Sep 18, 2008  Kaplan Allen D  Multiprotocol telecommunications routing optimization 
US20090285099 *  Nov 19, 2009  Colin Kahn  Method and apparatus for providing congestion control in radio access networks  
US20090296613 *  Jun 3, 2008  Dec 3, 2009  Colin Kahn  Method and apparatus for providing qualityofservice in radio access networks 
US20100046538 *  Aug 25, 2008  Feb 25, 2010  International Business Machines Corporation  Tcp connection resource diversity using tunable geometric series 
US20100080123 *  Sep 30, 2008  Apr 1, 2010  Colin Kahn  Method and Apparatus for Signaling Proprietary Information Between Network Elements of a Core Network in a Wireless Communication Network 
US20100080153 *  Apr 1, 2010  Colin Kahn  Method and apparatus for prioritizing packets for use in managing packets in radio access networks  
US20100153555 *  Dec 15, 2008  Jun 17, 2010  At&T Intellectual Property I, L.P.  Opportunistic service management for elastic applications 
US20110149794 *  Dec 20, 2010  Jun 23, 2011  Electronics And Telecommunications Research Institute  Apparatus and method for dynamically sampling of flow 
US20120151078 *  Dec 8, 2010  Jun 14, 2012  Sarat Puthenpura  Method and apparatus for capacity dimensioning in a communication network 
US20120192230 *  Jun 30, 2011  Jul 26, 2012  Samsung Electronics Co., Ltd.  Adaptive audio/video streams proxy 
US20120253761 *  Oct 19, 2010  Oct 4, 2012  Nec Corporation  Available bandwidth estimating device 
US20130272299 *  Mar 13, 2013  Oct 17, 2013  Patentmarks Communications, Llc  Multiprotocol telecommunications routing optimization 
US20140169163 *  Dec 18, 2012  Jun 19, 2014  General Electric Company  Systems and methods for communication channel capacity change detection 
US20150023169 *  Jul 19, 2013  Jan 22, 2015  Cisco Technology, Inc.  System and Architecture to Optimize Video Traffic over Internet Protocol Networks 
U.S. Classification  370/230, 370/229, 370/235 
International Classification  H04L12/56 
Cooperative Classification  H04L12/5602 
European Classification  H04L12/56A1 
Date  Code  Event  Description 

Sep 25, 2000  AS  Assignment  Owner name: TELEFONAKTIEBOLAGET L M ERICSSON (PUBL), SWEDEN Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:TELEK, MIKLOS;REEL/FRAME:011158/0973 Effective date: 20000828 Owner name: TELEFONAKTIEBOLAGET L M ERICSSON (PUBL), SWEDEN Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:RACZ, SANDOR;REEL/FRAME:011133/0921 Effective date: 20000822 
Sep 26, 2000  AS  Assignment  Owner name: TELEFONAKTIEBOLAGET L M ERICSSON (PUBL), SWEDEN Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:FODOR, GABOR;REEL/FRAME:011158/0321 Effective date: 20000906 
Mar 7, 2008  FPAY  Fee payment  Year of fee payment: 4 
Mar 7, 2012  FPAY  Fee payment  Year of fee payment: 8 
Mar 7, 2016  FPAY  Fee payment  Year of fee payment: 12 